feat(sema): lower natspec

[0xrusowsky]

Motivation

towards full natspec support; continues the work started in:

• feat(ast): naive natspec #470

[codspeed-hq]

CodSpeed Performance Report

Merging #567 will degrade performances by 5.81%

_{Comparing 0xrusowsky:rusowsky/lower-natspec (906c10b) with main (6006669)}

Summary

❌ 2 regressions
✅ 33 untouched

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
❌	UniswapV3/lower	4.8 ms	5 ms	-5.04%
❌	Vm/lower	2.8 ms	2.9 ms	-5.81%

[0xrusowsky]

still wip, currently working on a rewrite to validete and resolve inheritdoc in a single pass

[0xrusowsky]

@DaniPopes after validating all tags and resolving inheritdoc in a single pass, i managed to reduce the performance overhead on univ3/lower from ~20% to ~5%

unfortunately, that was useless on solady, as it doesn't have any @inheritdoc tags in the codebase... which means that what we want to optimize is the lowering/validation of the other tags.

i've tried several things like:

• hinting the compiler which fns to inline + which should be cold
• if also tried doing some lazy initialization for params/returns lookups
• i was already sorting match arm variants according to (what i believe to be) hot paths, with @notice first, and @dev second.

but none of that proved to be useful, as solady is always at around 24% worse than without lowering :(

i acknowledge that solady has a bunch of comments, and that this must have some runtime overhead, but it kinda feels too much?

some things that i noticed, which maybe can help u come up with improvement ideas:

• most of the comments are @dev comments (which don't require validation)
• most of the comments are line comments, which means that:
  1. we don't group them
  2. multiline comments create new natspec items with the implicit @notice tag (don't require validation either)

lmk what else should i try to improve perf

[DaniPopes]

Please look at codspeed at what's slower. It's because it emits a bunch of diagnostics

[0xrusowsky]

i was under the impression that there could only be a single @author tag, but it turns out that solc allows multiple of them.

this was causing a bunch of emitted diagnostics in solady as u pointed out

perf regression went down to -5.8% on Vm.sol, solady is not an issue anymore, which makes much more sense!

thanks @DaniPopes